Agentic data science with the Data Agent Kit
cloud.google.com·6h ago
AI Research
arXiv CS.AI·Aug 24·5 min read
ARQ: Agentic CodeQL Query Refinement for C/C++ Vulnerability Detection
TL;DR
ARQ is a novel framework that enhances CodeQL queries for detecting vulnerabilities in C/C++ code by automatically refining them based on execution evidence. It significantly reduces false positives and negatives, improving detection rates by up to 119.8% while maintaining high precision.
✦ Why It Matters
Implement ARQ to enhance your CodeQL queries and improve vulnerability detection in your C/C++ projects today.
Key Takeaways
- Use ARQ to refine CodeQL queries for better vulnerability detection.
- Expect up to 119.8% more true positives with ARQ-refined queries.
- Maintain a precision of at least 98.0% in vulnerability detection.
- Leverage ARQ to resolve existing issues in CodeQL queries.
How It Works
ARQ refines CodeQL queries by analyzing synthesized C/C++ programs to identify discrepancies between the program's execution and the query's results. When a synthesized program is vulnerable but the query does not flag it, this indicates a false negative.
Conversely, if the query flags a safe program, it indicates a false positive. ARQ uses these insights to iteratively improve the queries through a large language model (LLM) refinement loop.